<div dir="ltr">+1 for this being opt in. It's a substantial hit to build time<br></div><br><div class="gmail_quote"><div dir="ltr">On Thu, Aug 23, 2018 at 12:21 PM Christopher Larson <<a href="mailto:kergoth@gmail.com">kergoth@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div dir="ltr">IMO PGO should really stay opt-in, not opt-out, given the performance impact.<br></div><br><div class="gmail_quote"><div dir="ltr">On Thu, Aug 23, 2018 at 1:28 AM Richard Purdie <<a href="mailto:richard.purdie@linuxfoundation.org" target="_blank">richard.purdie@linuxfoundation.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On Wed, 2018-08-22 at 18:44 -0700, Andre McCurdy wrote:<br>
> On Thu, Aug 16, 2018 at 9:48 PM, Anuj Mittal <<a href="mailto:anuj.mittal@intel.com" target="_blank">anuj.mittal@intel.com</a>><br>
> wrote:<br>
> > On 08/17/2018 03:31 AM, Andre McCurdy wrote:<br>
> > > On Wed, Aug 15, 2018 at 11:26 PM, Anuj Mittal <anuj.mittal@intel.<br>
> > > com> wrote:<br>
> > > Is it important to re-create the profile data during every build<br>
> > > or<br>
> > > would we get most of the same benefits from using reference data<br>
> > > which<br>
> > > is generated offline?<br>
> > <br>
> > We should get the same benefit using the data generated offline as<br>
> > long<br>
> > as the source code/configure options/flags are same I believe. I<br>
> > have<br>
> > only tried with data generate offline using the same build<br>
> > configuration<br>
> > though.<br>
> <br>
> As an additional data point, python do_compile now takes approx 18<br>
> minutes on my laptop. A single qemu-i386 thread loading one CPU core<br>
> at 100%, with nothing else being scheduled in parallel for most of<br>
> that time.<br>
> <br>
> If we can get most of the benefits of pgo with a pre-generated data<br>
> file then that might still be something to explore.<br>
<br>
I've also noticed the increased build times. I'd love to see if we<br>
could create an sstate object which would contain the pgo data and<br>
which didn't depend on the usual compiler/toolchain pieces, only the<br>
do_patch stage checksum. In theory, the same sstate object should be<br>
reusable between target/native/nativesdk too as long as we had the same<br>
python patchset (which I know we don't for native).<br>
<br>
Anyone fancy such a challenge? We should probably have a bug open for<br>
this at least...<br>
<br>
Cheers,<br>
<br>
Richard<br>
<br>
-- <br>
_______________________________________________<br>
Openembedded-core mailing list<br>
<a href="mailto:Openembedded-core@lists.openembedded.org" target="_blank">Openembedded-core@lists.openembedded.org</a><br>
<a href="http://lists.openembedded.org/mailman/listinfo/openembedded-core" rel="noreferrer" target="_blank">http://lists.openembedded.org/mailman/listinfo/openembedded-core</a><br>
</blockquote></div><br clear="all"><br>-- <br><div dir="ltr" class="m_4532180818432277067gmail_signature" data-smartmail="gmail_signature"><div dir="ltr"><div><div dir="ltr">Christopher Larson<br>kergoth at gmail dot com<br>Founder - BitBake, OpenEmbedded, OpenZaurus<br>Senior Software Engineer, Mentor Graphics</div></div></div></div>
-- <br>
_______________________________________________<br>
Openembedded-core mailing list<br>
<a href="mailto:Openembedded-core@lists.openembedded.org" target="_blank">Openembedded-core@lists.openembedded.org</a><br>
<a href="http://lists.openembedded.org/mailman/listinfo/openembedded-core" rel="noreferrer" target="_blank">http://lists.openembedded.org/mailman/listinfo/openembedded-core</a><br>
</blockquote></div>